Module scaling 

Available on crate feature scaling only.

Scaling pressure calculation for autoscaler integration.

Produces a 0.0-100.0 composite metric based on weighted application signals with two hard gates (circuit breaker, memory pressure). Designed for KEDA but works with any autoscaler that reads Prometheus gauges.

Architecture

App signals ──→ ScalingPressure ──→ {prefix}_scaling_pressure gauge
                 ├─ Gate: circuit breaker open → 0.0
                 ├─ Gate: memory ≥ threshold → 100.0
                 └─ Weighted composite → 0.0-100.0

Usage

1. Define components with weights and saturation points
2. Create ScalingPressure with base config + components
3. Update component values from your pipeline (lock-free)
4. Call calculate() when rendering Prometheus metrics

use hyperi_rustlib::scaling::{ScalingPressure, ScalingPressureConfig, ScalingComponent};

let pressure = ScalingPressure::new(
    ScalingPressureConfig::default(),
    vec![
        ScalingComponent::new("kafka_lag", 0.35, 100_000.0),
        ScalingComponent::new("buffer_depth", 0.25, 10_000.0),
        ScalingComponent::new("memory", 0.40, 1.0),
    ],
);

// Update from pipeline (lock-free, call from any thread)
pressure.set_component("kafka_lag", 50_000.0);
pressure.set_memory(400_000_000, 1_000_000_000);

// Render in Prometheus endpoint
let value = pressure.calculate();
assert!(value >= 0.0 && value <= 100.0);

CPU Scaling

CPU is intentionally not included in the composite. KEDA’s native CPU trigger reads from the Kubernetes metrics-server (container-level CPU utilisation). Configure both triggers independently in your KEDA ScaledObject:

• scaling_pressure gauge → Prometheus scaler (app-level signals)
• CPU utilisation → CPU scaler (container-level, via metrics-server)

KEDA scales to the MAX of all triggers.

Structs

ComponentSnapshot

Per-component diagnostic snapshot.

PressureExpr

A single named pressure expression -> one {ns}_scaling_pressure{name=...} gauge. The autoscaler scales to the MAX across all enabled pressures.

PressureSnapshot

Full diagnostic snapshot of scaling pressure state.

PressureTargets

Per-pod normalisation targets (KEDA lagThreshold-style: “what ONE pod tolerates”). Pulled from the config params map with researched defaults.

RateWindow

Sliding window rate calculator.

ScalingComponent

Named scaling component with weight and saturation point.

ScalingEngine

The CEL-over-local-metrics horizontal scaling-pressure engine.

ScalingEngineConfig

Configuration for the horizontal scaling-pressure ENGINE (CEL over local metrics).

ScalingPressure

Lock-free scaling pressure calculator.

ScalingPressureConfig

Base configuration for scaling pressure calculation.

ScalingSignalsCell

Lock-free cell the app / transport updates with the current per-pod signals; the engine tick reads a snapshot. A NaN bit-pattern means “absent” -> None in the snapshot (contributes 0 to the pressure).

ScalingTransportConfig

Optional explicit transport kinds for the compound pressure (kafka, redis, http, grpc, …). None => auto-derived by the runtime.

TransportSignals

Per-pod, locally-knowable scaling signals, sampled each scaling tick.

Enums

GateType

Active gate preventing normal composite calculation.

ScalingTransport

Inbound/outbound transport kind, for scaling-signal selection.

Functions

inbound_pressure

Compound INBOUND pressure ratio (per-pod; >=0, unclamped above 1.0 for proportional scale-up). Picks the signal by the configured inbound kind.

outbound_pressure

Compound OUTBOUND pressure ratio (per-pod). EMIT-ONLY by default – NOT in the smart-default composite (downstream-bound; more pods rarely relieve a saturated sink – scaling ACR). A dead sink surfaces as the circuit gate in the composite, not here.